Study On Geometrical Structure Retrieval And Polarization Radar Recognition Of Man-made Targets

Man-made targets,such as tank,armored vehicle and Unmanned Aerial Vehicle(UAV),are the main objects of reconnaissance and strike on the battlefield.So recognizing man-made targets has great martial value.It is helpful for recognizing the target to extract geometrical structure of the man-made target based on SAR/ISAR image and characterize the contour of the target by the canonical geometrical structure.Therefore,geometrical structure retrieval is important for image interpretation and target recognition.This dissertation focuses on the geometrical structure retrieval and polarization recognition techniques for man-made targets,which are advanced issues,and deeply studies the spatial polarization characteristics of the canonical geometrical structures,geometrical structure retrieval technology and polarization recognition technology based on lots of fully polarimetric data and electromagnetic simulation data.The main contents are as follows:First of all,the spatial polarization characteristics of canonical geometrical structure are studied and the concept of Consistent-Polarization Angle Domain(CPAD)is put forward.The model of canonical geometrical structure is compared with the scattering field of the structure calculated by the electromagnetic simulation software.And the results show that the existing model has shortcomings in describing the polarization characteristics of the geometrical structure.Then,the spatial polarization characteristics of canonical geometrical structure are analyzed in detail and the concept of CPAD is proposed.The CPAD is used to describe the angle domain,within which the relation between the canonical geometrical structure and scattering mechanism is stable.The CPAD of the canonical geometrical structure is studied and the angle domain is determined,within which the scattering mechanism of canonical geometrical structure is the theoretical value.The study results support the image interpretation and automatic target recognition.The polarization characteristics of the close canonical geometrical structures are studied and the characteristic differences between close canonical geometrical structures and a single canonical geometrical structure are revealed.Based on these differences,the scattering center corresponding to a single canonical geometrical structure can be determined.Secondly,the geometrical structure retrieval method is studied,which retrieves the type and size of the local structure of the target using the scattering center classification and the clustering algorithm.The method for determining the canonical scattering center is proposed,which determines the scattering center corresponding to a single canonical geometrical structure based on the characteristics of echo amplitude persistent and polarimetric persistent.This method improves the accuracy of the scattering center classification.Then,based on the study of scattering center classification the clustering-based geometrical structure retrieval method is proposed,which can effectively retrieve the geometrical parameter,such as type and size,of the local structure of the target.The proposed method reduces the computation complexity compared with the model-based geometrical structure retrieval method and avoids model order selection and scattering center discrimination,which enlarges the application range.Thirdly,the polarization characteristics and the local structure characteristics of UAV with complex materials are studied.The geometrical structure of the UAV is retrieved to study the local structure characteristics.And the polarization characteristics of the UAV are studied based on the UAV data measured in anechoic chamber to reveal the differences in polarimetric scattering between the UAV and its perfect electric conductor model.The study results provide theoretical guidance and feature support for recognizing the target with complex materials.Finally,two efficient recognition algorithms are proposed and the experimental results on the real data demonstrate the feasibility and accuracy of the proposed methods.A feature vector with strong separability is designed,which can represent the structure characteristics of the target.And the fast density search clustering method is used to realize polarization recognition for man-made target.Then the polarization recognition algorithm based on model matching and pre-classification is proposed.Compared with the recognition method based on template matching,the proposed method avoids establish of template library and has the advantages of small storage capacity,fast calculation and wide application range.At the same time,the method reduces computation cost and improves the recognition accuracy by pre-classification which utilize the geometrical structure characteristics of the target.